News & analysis, rated.
Every story from the daily brief, each rated so you know what matters. Hand-picked by the desk, not an automated firehose.
Browse the news archive →OpenAI and Broadcom unveil Jalapeño, their first custom LLM inference chip
On Wednesday, OpenAI unveiled its first custom-built inference processor, designed and manufactured in collaboration with Broadcom; the chip, named Jalapeño, was designed specifically for the unique needs of OpenAI's inference systems. It is the first step in a multi-generation compute platform designed for initial deployment by the end of 2026. The companies say the chip reached tape-out in just nine months. It reportedly cuts inference costs by about 50%, according to Bloomberg. Coverage in TechCrunch, CNBC, and VentureBeat; see also the OpenAI announcement.
Anthropic accuses Alibaba of a 25,000-account campaign to distill Claude
Anthropic accused Alibaba of waging a large-scale effort to "illicitly" access its Claude model using thousands of fraudulent accounts, in a letter sent to several US senators and White House officials saying a campaign by operators linked to Alibaba's Qwen AI lab targeted Claude's most prized capabilities, including software engineering and agentic reasoning. The campaign involved nearly 25,000 fraudulent accounts and 28.8 million exchanges with Claude between April 22 and June 5, targeting capabilities including software engineering, agentic reasoning and long-horizon task planning , according to Bloomberg and a parallel Wall Street Journal report.
FDA grants first generative-AI radiology tools breakthrough device status
Cognita CXR is a generative vision-language model designed to assist radiologists in the interpretation of chest X-rays and is the first radiology generative AI model, and one of very few radiology AI solutions, to be granted FDA Breakthrough Device Designation. The designation was announced by Mosaic Clinical Technologies earlier this year and re-surfaced this week alongside a parallel designation for Aidoc, with internal validation data reporting detection gains of 16–65% across certain findings and an 18% interpretation efficiency boost.
Backbase buys Kasisto to embed agentic AI in its banking OS
Backbase announced on June 23, 2026 the acquisition of Kasisto, a pioneer in agentic AI for banking and financial services; Kasisto's agentic platform, financial services intelligence, and New York-based team are now part of Backbase and the AI-native Banking OS. The financial terms of the deal have not been made public. Kasisto counts JPMorgan Chase, TD Bank and Westpac among the 47 clients for its banking-specific conversational AI, and has raised tens of millions of dollars in funding, with investors including FIS, Mastercard and Westpac. Reporting from Finextra, PYMNTS, and Fintech Futures.
Google folds computer use into Gemini 3.5 Flash as a built-in tool
Google said computer use is now a built-in tool supported in Gemini 3.5 Flash, delivering its best performance yet for agentic computer use tasks; previously only available as a standalone Gemini 2.5 computer use model, computer use is now integrated natively in the main Gemini Flash model. Gemini 3.5 Flash achieves a 78.4 score on OSWorld-Verified tasks, matching premium models like Sonnet 4.6 while drastically outperforming Gemini 3 Flash at 65.1. See the DeepMind announcement and Neowin coverage.
Anthropic plants Claude inside Slack channels as a shared teammate
Anthropic on Tuesday launched Claude Tag, a shared AI agent embedded directly inside Slack, available in beta and research preview for Claude Enterprise and Team customers , replacing the older Claude in Slack app. Migration requires an administrator opt-in within 30 days, and Anthropic is issuing introductory launch credits to eligible Enterprise and Team organizations. The product runs on Opus 4.8 and operates under a single channel-scoped org identity rather than per-user permissions. Anthropic says 65% of its own product team's code is currently generated by an internal version of Claude Tag, and the pattern is also spreading to data analysis and support ticket resolution . Coverage in Techstrong.ai and IT Pro corroborate the launch detail and ambient-mode controls.
GPT-5 Pro cracks a three-year T-cell mystery in an immunology lab
OpenAI published a case study describing how immunologist Derya Unutmaz used GPT-5 Pro in late 2025 to revisit a three-year-old puzzle centered on a special type of immune cell that helps the body fight cancer and other illnesses . Unutmaz asked GPT-5 to predict the outcome of an experiment he had already conducted on a CD8+ T cell that targets lymphoma; the model correctly predicted the boost in the cells' ability to kill lymphoma cells, and it could not have gleaned the result from the internet because Unutmaz had not yet published it. OpenAI's writeup is the primary source; secondary analysis at Windows Forum notes Unutmaz works at The Jackson Laboratory for Genomic Medicine.
GitHub Copilot opens agent sessions to any model provider via BYOK
The GitHub Copilot app now supports bring your own key (BYOK), so you can run agent sessions against your own model providers, including OpenAI, Azure OpenAI, Microsoft Foundry, Anthropic, LM Studio, Ollama, and any OpenAI-compatible endpoint. Keys are stored in the local OS keychain and are never read back by the UI , per the GitHub changelog. The same release notes flag that Copilot CLI's new terminal interface is generally available .
Workday must face California AI-bias claims after dismissal denied
U.S. District Judge Rita F. Lin rejected Workday's attempt to dismiss claims that it violated the California Fair Employment and Housing Act and the Americans with Disabilities Act, finding the plaintiffs plausibly alleged enough of a California nexus for FEHA claims even for applicants outside California, because Workday allegedly designed, operated and controlled its AI hiring tools from California. HR Dive reports the judge was unmoved by Workday's argument that FEHA should not apply because many class members are non-residents, and Bloomberg Law notes the court also allowed a separate ADA disability claim to proceed.
MoEngage buys Aampe to put one AI agent behind every customer
Indian customer engagement software firm MoEngage acquired San Francisco-based startup Aampe in an all-cash deal, gaining technology that assigns AI agents to individual customers , per TechCrunch. Terms were not disclosed, but a source familiar with the transaction told TechCrunch the deal was worth tens of millions of dollars. Aampe's agents are already in production at Grab, Swiggy, ZenBusiness, and Taxfix, where the platform runs hundreds of millions of dedicated AI agents and processes more than 200 billion decisions every week .
NVIDIA says its Rubin AI data centers hit zero water consumption with closed-loop liquid cooling
NVIDIA's Sunday blog post calls its Rubin reference design the first to hit 100% liquid cooling — every chip in a closed loop, no fans. Director Ali Heydari says the DSX design has "zero water consumption." Independent coverage estimates it cuts on-site cooling water from ~2.6M gallons per megawatt a year to near zero in favorable climates.
OpenAI ships GPT-5.5-Cyber and Patch the Planet to automate vulnerability fixes at scale
On Monday OpenAI expanded its Daybreak program with the full release of GPT-5.5-Cyber, a Codex Security plugin, and Patch the Planet, an open-source patching effort with Trail of Bits and HackerOne. The model scored a record 85.6% on CyberGym, and launch partners include CrowdStrike, Palo Alto Networks, IBM and Wiz.